2007-02-22
SP2 SupportWP 2.1 Track bed quality assessmentTask 2.1.10 Numerical modelling of poor quality sites
First phase report on the modelling of poor quality sites
CZECH TECHNICAL UNIVERSITY IN PRAGUE
2008-02-21
2007-02-22
Numerical modelling: list of FE models
List of the studied models:
1. axisymmetric FE model of the experimental box
2. 3-D FE model of the experimental box
3. axisymmetric FE model of the in-situ conditions
4. 3-D FE model with reinforcing geogrid
2008-02-21
2007-02-22
FE modelling of laboratory experiments
1. Why? To reduce number of experiments needed to evaluate all possible situations
2. Variable parameters (material properties, construction layer thickness, contact behaviour)
3. To assess the importance of each variable observed (sensitivity analysis)
4. Result: nomograms (e.g. layer thickness vs. bearing capacity required)
2008-02-21
2007-02-22
5 strain gauge rosettes
from principal strains
using correlation method
with FE model calculate
the vertical
displacements
very precise assessment
of the deformation area
validation of the
numerical model
Settlement of the rubber (soil) measured by strain gauges
2008-02-21
2007-02-22
Results for 22.5 tons 25.0 tons and 27.5 tons loading
temperature compensation included
good correspondence with theory
deformation known in all points
2007-02-22
FEM model of the experimental box with sleeper
experimental box modelled
vertical displacement in good
correlation with the measured values
known strains and stresses
relationship between vertical
displacement and principal strains
2008-02-21
2007-02-22
Results - vertical displacements calculated vs. measured
- for all cases: B35/SB20/E20, B35/SB20/E20, B45/SB20/E20
- good correspondence with the experimental results for all
cases (average standard deviation is 16.7 %), except for
the case of ballast thickness 350 mm
- discrepancy caused by an error in measurement of the
sleeper deflections
- calculated displacements at the ballast sub-ballast
interface in good agreement with experimental values
even in the case of ballast thickness 350 mm (average
standard deviation less than 15.6 %)
2008-02-21
2007-02-22
Design graphs for single layer construction
axisymmetrical, fully parametric FE
models
results in terms of design graphs
horizontal axis = modulus of
deformation of the existing subgrade
vertical axis = sub-ballast thickness
required to achieve specified modulus
of deformation on the top
2007-02-22
Example of the design graphs
Required modulus
of deformation
<20 MPa,80 MPa>
Steps in 5 MPa
increments
2007-02-22
Design graphs for two layer construction
again axisymmetrical, fully parametric
FE models
results in terms of design graphs
thickness of the top layer held
constant,
design graphs for required modulus of
deformation in <50MPa, 90MPa>
2008-02-21
2007-02-22
Design graphs for construction with geosynthetics
again axisymmetrical, fully parametric
FE models
material properties of geosynthetics
important
proper modelling of the contact
studied
calibrated with the experimental
measurements from the box
2008-02-21
2007-02-22
Extension of our parametric FE models to rail track
plain strain and 3-D models
quadratic elements used
high quality of elements required
proper contact modelling studied
evaluation of design criteria
assessed using parametric modelling
2008-02-21
2007-02-22
results as design graphs
On-going work :: reinforcing effect of cement layer
2008-02-21
2007-02-22
Geogrids and geotextilies modelling in 3D
• 2-D and 3-D models
• importance of proper geogrid modelling
• proper contact behaviour modelling
• significance of different size of mesh
• contact behaviour and increase of
bearing capacity verified by physical
modelling in the experimental box
2007-02-22
Probabilistic approach
• scatter of input parameters (material properties)
described by e.g. Gaussian distribution with
standard deviation of ±3 – 5%
• how large is the scatter of the output parameters
(settlement, stress, etc)?
• which input variables contribute the most to the
scatter of an output parameter and to the failure
probability?
• what are the sensitivities of the output parameter
with respect to the input variables?
• Monte-Carlo, Latin hypercube, ...
2008-02-21